Background: The optimal revascularisation strategy for patients with left main coronary artery disease is uncertain. We therefore aimed to evaluate long-term outcomes for patients treated with percutaneous coronary intervention (PCI) with drug-eluting stents versus coronary artery bypass grafting (CABG). Methods: In this individual patient data meta-analysis, we searched MEDLINE, Embase, and the Cochrane database using the search terms “left main”, “percutaneous coronary intervention” or “stent”, and “coronary artery bypass graft*” to identify randomised controlled trials (RCTs) published in English between database inception and Aug 31, 2021, comparing PCI with drug-eluting stents with CABG in patients with left main coronary artery disease that had at least 5 years of patient follow-up for all-cause mortality. Two authors (MSS and BAB) identified studies meeting the criteria. The primary endpoint was 5-year all-cause mortality. Secondary endpoints were cardiovascular death, spontaneous myocardial infarction, procedural myocardial infarction, stroke, and repeat revascularisation. We used a one-stage approach; event rates were calculated by use of the Kaplan-Meier method and treatment group comparisons were made by use of a Cox frailty model, with trial as a random effect. In Bayesian analyses, the probabilities of absolute risk differences in the primary endpoint between PCI and CABG being more than 0·0%, and at least 1·0%, 2·5%, or 5·0%, were calculated. Findings: Our literature search yielded 1599 results, of which four RCTs—SYNTAX, PRECOMBAT, NOBLE, and EXCEL—meeting our inclusion criteria were included in our meta-analysis. 4394 patients, with a median SYNTAX score of 25·0 (IQR 18·0–31·0), were randomly assigned to PCI (n=2197) or CABG (n=2197). The Kaplan-Meier estimate of 5-year all-cause death was 11·2% (95% CI 9·9–12·6) with PCI and 10·2% (9·0–11·6) with CABG (hazard ratio 1·10, 95% CI 0·91–1·32; p=0·33), resulting in a non-statistically significant absolute risk difference of 0·9% (95% CI −0·9 to 2·8). In Bayesian analyses, there was an 85·7% probability that death at 5 years was greater with PCI than with CABG; this difference was more likely than not less than 1·0% (<0·2% per year). The numerical difference in mortality was comprised more of non-cardiovascular than cardiovascular death. Spontaneous myocardial infarction (6·2%, 95% CI 5·2–7·3 vs 2·6%, 2·0–3·4; hazard ratio [HR] 2·35, 95% CI 1·71–3·23; p<0·0001) and repeat revascularisation (18·3%, 16·7–20·0 vs 10·7%, 9·4–12·1; HR 1·78, 1·51–2·10; p<0·0001) were more common with PCI than with CABG. Differences in procedural myocardial infarction between strategies depended on the definition used. Overall, there was no difference in the risk of stroke between PCI (2·7%, 2·0–3·5) and CABG (3·1%, 2·4–3·9; HR 0·84, 0·59–1·21; p=0·36), but the risk was lower with PCI in the first year after randomisation (HR 0·37, 0·19–0·69). Interpretation: Among patients with left main coronary artery disease and, largely, low or intermediate coronary anatomical complexity, there was no statistically significant difference in 5-year all-cause death between PCI and CABG, although a Bayesian approach suggested a difference probably exists (more likely than not <0·2% per year) favouring CABG. There were trade-offs in terms of the risk of myocardial infarction, stroke, and revascularisation. A heart team approach to communicate expected outcome differences might be useful to assist patients in reaching a treatment decision. Funding: No external funding.